Should I Take to Drink? The Latest Evidence

Yet another area of human endeavor you can learn to enjoy!

These studies remain close to useless as long as they continue to lump together people who have never been drinkers with people who used to drink but have now stopped. The latter group includes people who drank to extreme excess, had medical or other problems as a result, and recently decided to stop.

It seems likely that those who used to drink heavily but then stopped would have problems similar to heavy drinkers.

z: If you read anything done in the last five years, it corrects for it. Start with Castelnuovo and Donati, 2006. Rimm and Moats 2007 is trenchant.

Those studies worried me too, in that I barely drink at all these days. But z has his suspicions and so do I. The reason I barely drink is that migraines get in the way, the headaches can get too bad. Which leads me to ask, how many of those who said they don't (or can't) drink stopped for similar reasons.

Look, I'd link to the studies, but I'm on crappy Christchurch earthquake internet. Rebecca: go read Castelnuovo and Donati - a huge metastudy that carefully separates out the ones that lump never and former drinkers. The separation very mildly attenuates the J curve, that's it.

Anti-alcohol crusaders love jumping up and down saying everything's contaminated by former drinkers. That might have been true ten years ago. But modern studies have fully tested it. Go check.

Rimm and Moats specifically test for the effect of former drinkers and use the harshest possible language for folks who continue to raise this possibility as reason for dismissing the J-curve.

It really makes social science look like a big, fat joke when smart people trained in statistics and econometrics can look at the findings of the best studies and go, "Yeah, but I still think the observed effects are probably due to something else."

I'm not saying that you're wrong to be skeptical. I'm saying that if it's OK to just throw away the best studies using large micro data sets and easily measured, well-defined variables, then how can we ever possibly get good enough data to answer any substantive policy questions, or anything in macro?

It seems like everyone is wasting their time.

It's even worse than that, Rapscallion. I've seen govt docs citing Rimm and Moats as "raising questions" about the J-curve due to confound of never-drinkers; the paper rather notes the question then refutes it!!

Intellectual dishonesty in this field is awful

It's even worse than that, Rapscallion. I've seen govt docs citing Rimm and Moats as "raising questions" about the J-curve due to confound of never-drinkers; the paper rather notes the question then refutes it!!

Intellectual dishonesty in this field is awful.

"However, even after adjusting for all covariates, abstainers and heavy drinkers continued to show increased mortality risks of 51 and 45%, respectively, compared to moderate drinkers"

It is plain nonsense, visible from airplane.

This post had me googling "Vitamin E fiasco". It seems entirely routine that epidemiology recommends interventions that go on to fail in randomised controlled trials. Is there anything here of interest to the RCT purist?

I'm not a 100% RCT purist. I still have hopes that the causal inference techniques being pioneered by Pearl, but I not seeing any mention of posteriori distributions over Bayesian networks. When researchers say that they "control" for this and that are they slogging away with the same old statistical techniques that don't work in theory and don't work in practise?

Life insurance companies have used moderate alcohol intake as a positive survival factor for decades. While severe drinking can clearly cause liver problems, post mortem analysis of alcoholics will usually show very clean arteries. Patients with severe heart problems back in the 70s were encouraged to drink.

These studies are not new. I would like to see the std deviations as well as the mean life expectancies with confidence intervals. The probabilty distribution overlap among the 3 groups is likely large.

Plus any study that is not controlled is really hard to get right. The statistics are not the problem per se, but the judgement which goes into "controlling for variables".

My best guess is go with the insurance companies.